System for monitoring and controlling a dynamic network

Abstract

The invention relates to a system for monitoring and controlling a dynamic network such as an oil, gas, or water pipeline. The system includes a plurality of sensors for measuring aspects of a state of the network with each sensor being associated with a segment of the network and connected to a virtual sensor which accumulates and pre-processes measurements from the sensors for each segment of the network. The system further includes a network topology processor for storing the topology of the network and relating sensors and virtual sensors to segments of the network and neighbouring sensors and virtual sensors in accordance with the topology and a reinforcement learning artificial neural network (ANN) based nonlinear state estimation and predictive control model which uses measurements from the sensors and virtual sensors to model the state of the network and estimate sequential states of the network.

The invention claimed is: 
     
       1. A system for monitoring and controlling a dynamic network comprising:
 a plurality of sensors for measuring aspects of a state of the network; each of the plurality of sensors being associated with a segment of the network and connected to a virtual sensor which accumulates and preprocesses measurements from the sensors for each segment of the network; 
 a network topology storing the topology of the network and relating sensors and virtual sensors to segments of the network and neighboring sensors and virtual sensors in accordance with the topology; 
 a reinforcement learning artificial neural network (RL-ANN) based nonlinear state estimation and predictive control model with measurements from the sensors and virtual sensors to model the state of the network and estimate sequential states of the network; 
 a classifier for classifying conditions of the segments of the network such that the condition of each segment may be monitored; 
 wherein preceding measurements from the sensors are temporarily stored and provided to the state estimation and predictive control model as inputs for modeling and estimating sequential states of the network; 
 wherein the ANN uses reinforcement learning to train the artificial neural network based on the difference between the estimated state and the measured state; 
 wherein the system includes a selective activation element which selectively provides and or withholds signals during the learning process; and 
 wherein selective activation element is utilized during training of the model to provide or withhold signals such that effective selections, where estimation error is smallest, are stored such that effective selections may be based on the stored values. 
 
     
     
       2. A system for monitoring and controlling a dynamic network comprising:
 a plurality of sensors for measuring aspects of a state of the network; 
 each of the plurality of sensors being associated with a segment of the network and connected to a virtual sensor which accumulates and preprocesses measurements from the sensors for each segment of the network; 
 a network topology processor for storing the topology of the network and relating sensors and virtual sensors to segments of the network and neighboring sensors and virtual sensors in accordance with the topology 
 a reinforcement learning artificial neural network (RL-ANN) based nonlinear state estimation and predictive control model which uses measurements from the sensors and virtual sensors to model the state of the network and estimate sequential states of the network; 
 a classifier for classifying conditions of the segments of the network such that the condition of each segment may be monitored; 
 wherein preceding measurements from the sensors are temporarily stored and provided to the state estimation and predictive control model as inputs for modeling and estimating sequential states of the network; 
 wherein the ANN uses reinforcement learning to train the artificial neural network based on the difference between the estimated state and the measured state; 
 wherein the system includes a selective activation element which selectively provides and or withholds signals during the learning process; and 
 wherein the selective activation element includes at least one synapse which encodes winning short-term memory activations into long-term memory. 
 
     
     
       3. The system of  claim 2  wherein the virtual sensors observe, identify, and learn inter-relationships of the measurements from the sensors of the segment which collectively represent a dynamic state of the network segment and behavior of the relevant segment. 
     
     
       4. The system of  claim 2  wherein the virtual sensors receive and process temporal physical measurements from the sensors such as pressure, flow, temperature, and computes derived measurements such as signal frequency, wavelength, autocorrelation, and signal mode. 
     
     
       5. The system of  claim 2  wherein network topology processor computes boundaries, divides the network into segments, and determines parameters of the segment. 
     
     
       6. The system of  claim 5  wherein the parameters include interconnections, length, number, and type of sensors which exhibit similar observable behavior. 
     
     
       7. The system of  claim 2  wherein the ANN is a dynamic ANN with dynamically configurable number of layers and neurons per layer. 
     
     
       8. The system of  claim 2  wherein the sensors include actuated sensors which measure the state of an actuator which acts upon the system such that the state estimation and predictive control model learns the effect of the state of the actuator on the state of the network. 
     
     
       9. The system of  claim 2  wherein actuators are controlled by the system and control of an actuator is based on the estimated states of the network by the state estimation and predictive control model and the desired state of the network. 
     
     
       10. The system of  claim 2 or claim 9  wherein the system includes a feature extraction element in the form of a recurrent neural network with memory, which uses temporal signals to build abstract profiles and extract repeatable phenomenon from the temporal wave behavior, its frequency elements and ability to propagate. 
     
     
       11. The system of  claim 2  wherein the system includes one or more preprocessing elements for preprocessing any measurements from virtual sensors for use by the active elements of the state estimation and predictive control model. 
     
     
       12. The system of  claim 11  wherein the network topology processor may normalize the measurements of virtual sensors, extract features from measurements and derived measurements, classify the extracted features which is provided to the state estimation model, and apply filters to the measurements to provide preprocessed values to the state estimation and predictive control model. 
     
     
       13. The system of  claim 2  wherein the classifier includes a Bayesian neural network for classifying the condition of each portion of the network. 
     
     
       14. The system of  claim 13  wherein the condition may be steady state, transitioning, or abnormal. 
     
     
       15. The system of  claim 2  wherein the classifier validates measurements from adjacent sensors to identify faulty readings. 
     
     
       16. The system of  claim 2  wherein the state estimation and predictive control model includes an anomaly detection element. 
     
     
       17. The system of  claim 16  wherein the anomaly detection element detects measurements or derived measurements which fall outside of the estimated state of the network. 
     
     
       18. The system of  claim 2  wherein the system includes a central controller which receives the measurements from the virtual sensors, stores historical measurements and estimated states, and provides measurements to the classifier and the state estimation and predictive control model. 
     
     
       19. The system of  claim 2  wherein the network topology processor stores the topology of the network as a graph. 
     
     
       20. The system of  claim 19  wherein the topology processor includes a geographical relation element which relates the nodes and vertices of the network to a physical location thereof. 
     
     
       21. The system of  claim 2  wherein the graph includes vertices, which represent nodes of the network, and edges, representing sections of the network. 
     
     
       22. The system of  claim 2  wherein the topology and geographical relation element correlate data structures that hold the spatial data, connectivity, real-time data, computed state space and model information of the network. 
     
     
       23. The system of  claim 2  wherein the system includes a fault detection element. 
     
     
       24. The system of  claim 23  wherein the fault detection element includes a geographical locating element, which uses the topological processor and the geographical relation element to locate the origin of the fault in the network. 
     
     
       25. The system of  claim 24  wherein the geographical locating element uses the geographical relation element, wave parameters, and triangulation to determine the physical location of the origin of the fault. 
     
     
       26. The system of  claim 23  wherein the fault detection element detects observable measured events which are measured at multiple locations which do not fall within the estimated state of the network and classifies the event as a fault. 
     
     
       27. The system of  claim 23  wherein the fault detection element detect measurements which correspond to a predefined fault condition. 
     
     
       28. The system of  claim 27  wherein the predefined fault condition includes water hammer, leak, and implosion in a section of the network. 
     
     
       29. The system of  claim 23  wherein the fault detection element includes a feature extraction element and a classifier. 
     
     
       30. The system of  claim 29  wherein the fault detection element includes a memory (LSTM) ANN which learns and extracts features across time steps in a window to detect faults which have a temporal aspect. 
     
     
       31. The system of  claim 29  wherein the fault detection element includes an input LSTM layer, multiple hidden layers, and output layer with softmax activation. 
     
     
       32. The system of  claim 2  wherein the system includes a graphical interface which shows the sections and nodes of the network. 
     
     
       33. The system of  claim 32  wherein an actuator is a pump and the sensor measures the operational level of the pump. 
     
     
       34. The system of  claim 32  wherein the graphical interface indicates the type and location of a fault. 
     
     
       35. The system of  claim 32  wherein the sections and nodes are shown schematically. 
     
     
       36. The system of  claim 32  wherein the sections and nodes are shown to scale. 
     
     
       37. The system of  claim 32  wherein the graphical interface indicates the condition of each section of the network. 
     
     
       38. The system of  claim 2  wherein the network is a pipeline network. 
     
     
       39. The system of  claim 38  wherein an actuator is a valve and the sensor measures the operational position of the valve. 
     
     
       40. The system of  claim 38  wherein the network is a water reticulation network. 
     
     
       41. The system of  claim 38  wherein the sensors include flow sensors, pressure sensors, temperature sensors, acoustic sensors, vibration sensors, turbidity sensors, pH sensors, Oxidation Reduction Potential (ORP) sensors, and chlorine sensors.